Various kinds of signal recording technologies are employed in many different kinds of instruments for the measurement of a time-varying voltage, such as x–y recorders, oscilloscopes, and analog-to-digital converters (ADC's) combined with memory arrays. In the latter configuration, the voltage waveform is recorded by sampling the voltage at well-defined, discrete time intervals with the ADC. The resulting digital values are transferred sequentially to storage elements of a memory array, creating a digital record of the voltage waveform, which may be subsequently processed and stored using a computer. If the voltage waveform is repetitive, then the same voltage waveform may be recorded similarly a number of times. By averaging the digitized waveforms, that is, by summing data values in all records that correspond to the same relative time-phase of the waveform, and dividing the result by the number of records, the signal-to-noise ratio of the measurement may be improved. Specifically, it is well known that, if the noise can be characterized as so-called random or white, Gaussian noise, then the root-mean-square (RMS) magnitude of the noise in the spectrum decreases in proportion to the square root of the number of waveform measurements that are averaged, while the signal remains the same, so that the signal-to-noise ratio improves accordingly.
A measurement instrument specially tailored to perform such signal averaging of repetitive voltage waveforms is conventionally referred to as a digital signal averager (DSA). A generic DSA typically includes an ADC; a precision clock for timing the intervals between consecutive analog-to-digital conversions, as well as timing the repetitive sequence of data transfer, processing and storage; an array of memory elements for storing and providing access for subsequent processing of the digitized data; and hardware for performing arithmetic operations on the data in the memory array elements; as well as hardware for performing other auxiliary functions such as data processing, data transfer, etc. Often, the signal averaging process is performed in real time. That is, as each voltage waveform is digitized, each digital data value output by the ADC is added promptly to the value in the memory array element that represents the summation of all previously recorded values of the voltage waveform at this same phase position in the waveform. After a desired number of waveforms have been summed, the resulting integrated waveform may be scaled to produce a true average waveform by dividing the summed waveform values by the number of waveform measurements records, although this extra operation, if performed at all, is often performed only after the summed waveform data is transferred out of the DSA and into computer memory for so-called post-processing. In principle, it is possible to envision the signal averaging, process being performed after all repeated measurements of the voltage waveform have been completed, whereby each waveform measurement had been recorded in a separate memory array or array segment. However, a large number of records is usually desired for signal averaging in practice, and the corresponding memory requirements usually, but not always, become impractical with such an approach. Hence, signal averaging is most commonly performed in real time, or approximately real time, as consecutive waveforms are recorded.
In order to signal average repetitive waveforms, it is necessary that the same phase relation is maintained among all recorded waveforms during the summation process. This simply means that the value in any particular memory location must be the summation of all repeated measurements of the signal at the same relative phase of the waveform. There are different schemes that may be employed to accomplish this, depending on the experimental situation. In one situation, it may be possible to detect a particular feature of the waveform of interest, and to trigger the start of the waveform recording process on such a feature; as is commonly done in digital oscilloscopes. In such situations, there will inevitably be an uncertainty in the relative phases between different recordings of the waveform of up to plus or minus one ADC digitization interval because of the lack of synchronization between the signal waveform and the ADC clock internal to the DSA.
Another common situation is that in which the voltage waveform to be measured may be produced on demand from the experiment in response to a trigger signal, and where the resulting voltage waveform is reproduced with the same phase relation to the trigger each time. In this case, it is advantageous, with respect to achieving the best timing accuracy and precision, to arrange for the trigger signal to be synchronous with the internal clock of the DSA. Then, the start of the waveform recording process by the DSA, commonly referred to as a ‘scan’, may be initiated at a time or phase relative to this so-called ‘Start’ signal that is the same for all recorded waveforms. Consequently, it is straightforward to ensure that the contents of any particular memory array location will always correspond to the signal at the same point on the waveform for each scan, resulting in the correct summation of consecutive scans. The complete signal waveform average of a number of scans is commonly referred to as a ‘record’. Digital signal averagers that have been designed for and employed in these latter experimental situations are disclosed, for example, in U.S. Pat. Nos. 4,490,806, 5,428,357, and 6,094,627. Examples of current state-of-the-art commercial implementations of such DSA's include the FastFlight DSA from EG&G Ortec, the Eclipse DSA from EG&G Signal Recovery, and the AP100/200 from Acqiris SA.
An illustration of this latter experimental situation is shown schematically in FIG. 1, which depicts a conventional DSA 30 being used to record the time-dependent signal 5 constituting a portion of interest of an ion mass spectrum produced by a time-of-flight mass analyzer 25. In the particular configuration of a time-of-flight mass analyzer 25 illustrated in FIG. 1, ions of a sample of interest 24 are introduced, by any of a variety of different methods that are well-known to those skilled in the art (such as laser desorption/ionization from a surface, orthogonal injection from an external ion source, electron-impact ionization of resident gas molecules, etc.) into a region of the mass analyzer known as the pulsing region 23. Upon application of a voltage pulse 20 from a voltage pulse generator 21 to an electrode 22 bounding the pulsing region 23, ions are accelerated into the time-of-flight measurement chamber 26, and arrive at a detector 27 with a time dependence that is proportional to the square-root of their mass-to-charge ratio. As ions arrive at the detector 27, a signal 5, related to the number of ions present for each mass-to-charge ratio, is generated, which is introduced, usually after amplification, to the input 6 of the DSA 30 for measurement and recording. Such time-dependent signal waveforms may therefore be interpreted in terms of the types of chemical species present in the original sample, as indicated by the various mass-to-charge peaks in the spectrum, while the amplitude of such peaks provide some measure of the relative amounts of the different species. A number of such scans are typically signal averaged to improve the signal-to-noise ratio of the measurement. The voltage pulse 20 that begins each time-of-flight spectrum measurement is triggered by a signal at the DSA trigger output 4 in synchronization with the internal clock of the DSA. Therefore, each scan may easily be synchronized with all previous scans, which is necessary to achieve a valid signal average record of all the scans.
An example of such a time-of-flight spectrum is illustrated in FIG. 2. This data was obtained from a commercially available time-of-flight mass spectrometer, manufactured and sold by Amersham Biosciences as the mass spectrometer of the Ettan LC/MS system. This system utilizes a customized version of the FastFlight DSA commercially available from EG&G Ortec. The data shown in FIG. 2 was acquired from a sample of hexatyrosine (20 pmole/uL in 50/50 methanol/water, 0.1% acetic acid), which was introduced into the electrospray ion source of the Ettan TOF mass spectrometer by constant infusion at a flow rate of 10 uL/min.
The spectrum of FIG. 2 is an average of 10 scans. The DSA signal offset setting was adjusted so that the entire signal waveform as well as the noise characteristics could be observed. In fact, the signal offset setting was adjusted so that the average zero-level baseline of the spectrum more or less coincided with 1 LSB of the ADC, as indicated by the dashed line in FIG. 2. Therefore, both positive and negative excursions from the average baseline, due to both noise and negative fluctuations in the signal (provided the amplitude of negative excursions does not exceed 1 LSB per scan), could be observed and measured. In the spectrum of FIG. 2, some of the ion signal peaks are apparent, for example, at flight times corresponding to memory array bins 9,992; 18,143; 25,553; and 33,935. Noise is also clearly evident throughout the spectrum of FIG. 2. Digital signal averaging is effective at improving the signal-to-noise ratio in such measurements of repetitive waveforms, but only if the noise is not coherent with the measured waveform, as with randomly distributed noise, or white Gaussian noise. Coherent noise, that is, noise that is time-correlated with the measured signal waveform, behaves just as the waveform signal does with respect to a signal averaging process. Consequently, owing to its coherency with the waveform signal, coherent noise is not reduced at all by conventional signal averaging. Hence, coherent noise is often the most important factor that limits the signal-to-noise ratio and dynamic range that can be achieved by a digital signal averager.
Noise that is coherent with the measured repetitive waveform may originate from sources coherent with the signal source, such as cable reflections, ringing due to impedance mismatches in the signal path, partial coupling of the initial trigger stimulus to the response signal, etc. Such sources of coherent noise can be made relatively insignificant with careful engineering design. However, coherent noise may also originate from within the digital signal averager itself, often due, for example, to coupling between the signal input and voltage transitions that occur internal to the DSA. Internal voltage transitions arise, for example, from read, add, and write operations that comprise the digital signal averaging process in the DSA electronics. Because the sequence of such voltage transitions are repeated precisely for each digitization, the noise that they generate at the signal input is repetitive and synchronous with the signal waveform being averaged.
In order to minimize coherent noise, a substantial effort is typically invested during the design and development of DSA's to isolate the analog-to-digital conversion stage, and associated analog circuitry, from the digital circuitry associated with transfer and processing of the digital data produced by the ADC. With careful engineering design, it is possible to reduce internally-generated coherent noise to a fraction of 1 least significant bit (LSB) of the ADC. For example, the aforementioned FastFlight DSA from EG&G specifies that the maximum internally-generated coherent (correlated) noise to be equivalent to <0.2 mV rms noise at the ADC input. Given that 1 LSB spans 3.9 mV of input signal voltage for this instrument, the maximum coherent noise of 0.2 mV rms corresponds to an rms noise of 0.051 LSB. In this case, unless steps were taken to suppress the coherent noise, the limit to the signal-to-noise ratio that could be achieved, due to this coherent noise, would be 255/0.051=5000 for this 8-bit DSA, regardless of the number of waveforms that were included in the signal average. However, in terms of the available signal dynamic range that this would imply, if the minimum detectable signal is defined to have a signal-to-noise ratio greater than 5, say, then the maximum signal dynamic range that is possible with this coherent noise would be limited to ˜1000. This assumes, of course, that enough waveforms are signal averaged so that any incoherent noise is rendered negligible.
Referring to the measurement results in FIG. 2, an illustration of the noise that is characteristic of that throughout the spectrum is shown in FIG. 3A, which portrays an expanded view of the portion of the spectrum from memory array bins 19,800 to 20,200. The noise illustrated in FIG. 3A appears to have some regularity, which is a signature of coherent noise. The measurement portrayed in FIG. 2 was repeated but with an average of 100 scans instead of 10. The noise in this spectrum, again as represented by contents of the memory array bins 19,800 to 20,200, is illustrated in FIG. 3B. The advantage of signal averaging in reducing the noise is apparent from a visual comparison of FIGS. 3A and 3B. In quantitative terms, the root-mean-square (RMS) of the noise of FIG. 3A is 0.32, while that of FIG. 3B is 0.27, a reduction by 16%. However, if this noise were random, or ‘white’ noise, then averaging 10 times more scans should theoretically reduce the noise by a factor equal to the square root of 10, or a factor of 3.33. Closer inspection of the noise in FIG. 3B clearly reveals a definite pattern, which is inconsistent with random noise, but which is characteristic of coherent noise, presumably originating from internal voltage transitions within the DSA itself.
Similarly, after-averaging 1000 and 10,000 scans, the noise in the same region of the spectrum is shown in FIGS. 3C and 3D, respectively. The coherent noise remains, and becomes more distinct, while the random component to the noise is reduced further, as expected, as more scans are signal averaged. The signal-to-noise ratio remains essentially the same as it was with a signal average measurement of 100 scans, due to the fact that the noise in the spectrum is dominated by the coherent noise, which is not reduced by conventional signal averaging techniques.
Various approaches have been devised in attempts to reduce coherent noise in DSA's. In the exemplary FastFlight DSA, for example, a constant voltage offset may be added to the signal input prior to the analog-to-digital conversion so that the coherent noise always falls below the minimum voltage corresponding to 1 LSB, provided that no other signal is present at the ADC input. With this approach, then, coherent noise may be eliminated in regions of the measured waveform that are very close to the baseline. However, the coherent noise is, nevertheless, fully manifest in regions of the waveform that have voltage amplitudes equal to or greater than the applied voltage offset, where the signal rises to the level of 1 LSB and above. Therefore, for signal levels in the measured waveforms that are at least as great as the applied constant voltage offset, this approach is unable to reduce the waveform distortions caused by coherent noise. Such waveform distortions due to coherent noise are especially problematic for relatively small signal levels, which are of amplitudes that are just great enough to correspond to the voltage level of 1, or several, LSB's. In these cases, the voltage excursions due to the coherent noise result in substantial noise on such small signal levels, which frustrates any attempt to obtain an accurate measurement, and limit the dynamic range of the DSA. Furthermore, such a voltage offset precludes the measurement of any signal with a magnitude less than that of the voltage offset, and therefore further limits the signal dynamic range capability of the DSA.
Another approach that has been used to reduce coherent noise in DSA's is that of noise filtering. It is well known to those skilled in the art that coherent noise in a DSA is sometimes composed of a limited and relatively well defined range of frequency components. It is sometimes advantageous, then, to apply a frequency filter, implemented either with hardware or with software, which is optimally tailored to reduce the frequency components of the coherent noise in the measured waveforms. Unfortunately, such filtering techniques unavoidably distort all features in the waveform, including the desired signal waveform characteristics, at least to some extent, and such distortion of the desired waveform is often unacceptable.
Still another approach to reducing coherent noise in DSA's is that of background subtraction. Essentially, this technique entails a measurement of the coherent noise spectrum without the presence of the signal waveform. Then, the measured coherent noise spectrum is subtracted from a measurement of a signal waveform, leaving, in principle, the signal waveform without coherent noise. This technique works well provided that the coherent noise spectrum is the same regardless of whether signal is present or not. Unfortunately, this is usually not the case. For example, for the FastFlight DSA from EG&G, the coherent noise is specified to be only a fraction of a least-significant-bit in magnitude. Therefore, without signal present, it is possible to adjust the zero-level offset to one extreme condition between two bit-transition boundaries, so that excursions from this zero-level due to the coherent noise never cross a digitizer bit transition, and, therefore, never appears in a measurement of the coherent noise spectrum. Alternatively, it is possible to adjust the zero-level offset to a voltage close to that of a bit transition, so that coherent noise produces excursions above and below the bit transition, resulting in a background spectrum that is characteristic of the coherent noise spectrum with this offset level. Clearly, the coherent noise that is manifest at any point in the spectrum of a measured signal waveform will depend strongly on the actual signal level that is present at that point in the signal waveform. Therefore, the subtraction of a background coherent noise spectrum from a signal waveform will only be effective for the portions of the signal waveform that are similar in amplitude, relative to a bit transition, as the zero-offset level with which the background coherent noise spectrum was measured, and therefore, such background subtraction is of limited utility. Even further, such background coherent noise subtraction may, in fact, result in an actual increase in coherent noise for some other portions of the measured signal spectrum, specifically, with amplitudes that correspond to voltage levels between two bit transitions, and, therefore, which would have appeared to be relatively free of coherent noise without the coherent noise background subtraction.
In summary, there has not been available a satisfactory solution to the reduction of internally-generated coherent noise in repetitive signal waveforms measured with digital signal averagers. Therefore, the signal-to-noise ratio and signal dynamic range that may be achieved with current state-of-the-art DSA's has been limited. The present invention described herein provides for the reduction of such coherent noise without any of the disadvantages of prior methods and apparatus.
Accordingly, it is one object of the present invention to provide devices and methods for the reduction of coherent noise in repetitive signal waveforms measured by digital signal averagers.
Another object of the present invention is to provide devices and methods for the reduction of coherent noise in repetitive signal waveforms measured by DSA's in real time, that is, during the signal waveform measurement and averaging process.
A still further object of the present invention is to provide devices and methods for the reduction of coherent noise in repetitive signal waveforms measured by DSA's that maintain the signal waveform fidelity for all signal levels.
Another object of the present invention is to provide devices and methods that are effective at reducing coherent noise in repetitive signal waveforms measured by DSA's at any setting of the voltage zero-offset level.
Yet another object of the present invention is to provide devices and methods for the reduction of coherent noise in repetitive signal waveforms measured by a DSA while maintaining the speed with which signal waveforms can be measured and signal averaged by the DSA.
Other objects and advantages over the prior art will become apparent to those skilled in the art upon reading the detailed description together with drawings as described as follows.